Improving Data Segmentation Accuracy Through Clinician-Informed Contextual Reasoning.

Area of Science:

• Health Informatics
• Clinical Data Management
• Data Privacy

Background:

• Current data segmentation technologies often fail to leverage crucial contextual information from electronic health records (EHRs).
• This limitation hinders accurate inferences for sensitive health data classification.

Purpose of the Study:

• To understand how physicians use contextual reasoning for classifying sensitive health data in EHRs.
• To inform the design of context-aware data segmentation tools.

Main Methods:

• Conducted semi-structured interviews with 24 board-certified physicians.
• Physicians categorized data from patient summaries and their confidence was assessed.
• Analyzed reasoning patterns by examining context types and data categories.

Main Results:

• 19 out of 24 physicians (79%) utilized contextual reasoning, referencing diagnoses, medical, and medication history.
• Medication history (38.9%) and laboratory results (36.1%) were most frequently categorized.
• High confidence was linked to structured data with diagnoses; ambiguity arose from social history.

Conclusions:

• Physicians routinely use contextual information for sensitive data, particularly for labs and medications.
• Integrating clinician reasoning into segmentation logic can enhance accuracy and privacy.
• Future research should explore generalizability and patient perspectives for equitable data sharing.